This could be a potential performance improvement. Basically, each Clifford gate rehashes every key in the map. This is wasted effort, since we know that applying a Clifford gate will never lead to collisions of existing keys. Rather, we'd only need to rehash before applying a non-Clifford gate or noise channel.
Not sure how large this gain would be, but it was significant enough in generalized tableau propagation to motivate disabling rehashing altogether (not needed there at all).
This could be a potential performance improvement. Basically, each Clifford gate rehashes every key in the map. This is wasted effort, since we know that applying a Clifford gate will never lead to collisions of existing keys. Rather, we'd only need to rehash before applying a non-Clifford gate or noise channel.
Not sure how large this gain would be, but it was significant enough in generalized tableau propagation to motivate disabling rehashing altogether (not needed there at all).